Aprendizado supervisionado incremental de redes bayesianas para mineração de dados

• Ano de defesa: 2007
• Autor(a) principal: Yoshida, Murilo Lacerda
• Orientador(a): Hruschka Júnior, Estevam Rafael
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de São Carlos
• Programa de Pós-Graduação: Programa de Pós-Graduação em Ciência da Computação - PPGCC
• País: BR
• Palavras-chave em Português: Inteligência artificial; Aprendizado incremental; Data mining (Mineração de dados); Representação do conhecimento
• Área do conhecimento CNPq: CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO
• Link de acesso: https://repositorio.ufscar.br/handle/20.500.14289/354

The objective of this work is to introduce two algorithms for supervised Bayesian network incremental learning, AIP (Algorithm for simple Bayesian network numerical parameters supervised incremental learning) and ABC (Algorithm for Bayesian network supervised incremental learning in layers). In order to develop these algorithms we studied relevant works about the Bayesian networks concepts, the algorithms for supervised Bayesian network learning and the algorithms for incremental supervised Bayesian network learning. To improve the performance of the ABC algorithm, we studied the AD-Tree structure and implemented it on the algorithm. To measure the quality of the networks learned by the algorithms we used these networks learnt to classify a test set, resulting in the correct classification rate (ICC). To do that we studied the test set classification process and the propagation of evidences along the Bayesian network. The result of the studies is described on this work, along with the results and discussions about the experiments made with the introduced algorithms.